Evaluation et analyse de sensibilite du modele CERES-Maize en conditions alsaciennes

CERES-Maize is a simulation model of maize growth and development, conceived and successfully used in the United States. In this study, CERES-Maize was tested under European conditions, then adapted to these conditions. Two field experiments were carried out for 2 years in North-Eastern France (Alsace) using 2 cultivars (LG11 and DEA) on a loamy soil. Yearly nitrogen fertilization was 150 kg.ha-1 (N). No irrigation was performed. Calibration and validation of the model were conducted for each cultivar, followed by sensitivity analysis of the newly calibrated model for the LG11 variety. Genetic constants, ie growing degree days (base 8°C) from seedling emergence to the end of the juvenile phase (P1), potential kernel number (G2) and potential kernel growth rate (G3), are generally unknown for hybrids grown in Europe (table I). Their values were therefore estimated from data given for early hybrids grown in Southern Canada and from French experiments testing DEA and LG11 (table II). Results showed that predictions given by the non-calibrated model using estimated constants are generally very different from measured values (table III). Therefore, genetic constants and equations calculating leaf area were modified (fig 2). Then, validations taken from other data sets than those used for calibration showed a good agreement between observed and predicted values for plant development and grain yield (tables IV and VI). Growth variables are still slightly underestimated by the model. This might be due to the fact that parameters used for calculating conversion of radiation in dry matter and repartition of dry matter between organs are not correct as regards European conditions. The sensitivity analysis revealed a great sensivity of genetic and hydraulic parameters (figs 3, 5 and 6). The sensitivity depends on simulation conditions, especially on the occurrence of water or nitrogen stress (fig 3). Other soil characteristic parameters were not found to be as sensitive as albedo (fig 4). CERES-Maize can be used for simulation of maize growth, development and for grain yield prediction in European conditions. However, a thorough calibration of the model and further experiments to accurately determine parameters unknown in Europe (eg end of the juvenile phase) are required. Furthermore, some subroutines should be included to take into account specific conditions such as cold springs and hydromorphic soils.

CERES-Maize est un modèle de simulation de la croissance et du développement du maïs élaboré et validé aux Etats-Unis. Afin d’estimer les possibilités de transposition du modèle dans des conditions européennes, des simulations ont été réalisées pour 2 variétés de maïs (LG11 et DEA) cultivées en Alsace 2 années consécutives. Pour chaque variété, le modèle a été calibré sur 1 année et validé sur la suivante. L’analyse de sensibilité des paramètres liés à la variété et au sol montre que la réponse du modèle varie fortement selon la nature du paramètre et les conditions agro-climatiques considérées dans l’analyse. En outre, les paramètres du modèle sont nombreux et non indépendants, ce qui rend le calibrage délicat sur le plan mathématique. D’un point de vue méthodologique, l’analyse de sensibilité de modèles tels que CERES-Maize et la généralisation du calibrage peuvent donc poser d’importants problèmes. Dans les conditions agro-climatiques des simulations (Alsace, sol limoneux, années 1978, 1979, 1986 et 1987), le modèle prédit correctement le développement et la croissance du maïs. L’ajustement de paramètres est alors nécessaire, mais la structure du modèle n’est pas remise en cause. Une généralisation à l’ensemble de l’Europe nécessiterait par contre la modification du modèle, en particulier pour la partie qui simule la phase d’implantation de la culture.